Aarch64: handle csel with icmp/fcmp source without materializing the bool.

Previously, we simply compared the input bool to 0, which forced the
value into a register (usually via a cmp and cset), zero-extended it,
etc. This patch performs the same pattern-matching that branches do to
directly perform the cmp and use its flag results with the csel.

On the `bz2` benchmark, the runtime is affected as follows (measuring
with `perf stat`, using wasmtime with its cache enabled, and taking the
second run after the first compiles and populates the cache):

pre:

       1117.232000      task-clock (msec)         #    1.000 CPUs utilized
               133      context-switches          #    0.119 K/sec
                 1      cpu-migrations            #    0.001 K/sec
             5,041      page-faults               #    0.005 M/sec
     3,511,615,100      cycles                    #    3.143 GHz
     4,272,427,772      instructions              #    1.22  insn per cycle
   <not supported>      branches
        27,980,906      branch-misses

       1.117299838 seconds time elapsed

post:

       1003.738075      task-clock (msec)         #    1.000 CPUs utilized
               121      context-switches          #    0.121 K/sec
                 0      cpu-migrations            #    0.000 K/sec
             5,052      page-faults               #    0.005 M/sec
     3,224,875,393      cycles                    #    3.213 GHz
     4,000,838,686      instructions              #    1.24  insn per cycle
   <not supported>      branches
        27,928,232      branch-misses

       1.003440004 seconds time elapsed

In other words, with this change, on `bz2`, we see a 6.3% reduction in
executed instructions.

cranelift/codegen/src/isa/aarch64/lower_inst.rs

@@ -1078,8 +1078,24 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
             // Nothing.
         }
 
-        Opcode::Select | Opcode::Selectif | Opcode::SelectifSpectreGuard => {
+        Opcode::Select => {
-            let cond = if op == Opcode::Select {
+            let flag_input = inputs[0];
+            let cond = if let Some(icmp_insn) =
+                maybe_input_insn_via_conv(ctx, flag_input, Opcode::Icmp, Opcode::Bint)
+            {
+                let condcode = inst_condcode(ctx.data(icmp_insn)).unwrap();
+                let cond = lower_condcode(condcode);
+                let is_signed = condcode_is_signed(condcode);
+                lower_icmp_or_ifcmp_to_flags(ctx, icmp_insn, is_signed);
+                cond
+            } else if let Some(fcmp_insn) =
+                maybe_input_insn_via_conv(ctx, flag_input, Opcode::Fcmp, Opcode::Bint)
+            {
+                let condcode = inst_fp_condcode(ctx.data(fcmp_insn)).unwrap();
+                let cond = lower_fp_condcode(condcode);
+                lower_fcmp_or_ffcmp_to_flags(ctx, fcmp_insn);
+                cond
+            } else {
                 let (cmp_op, narrow_mode) = if ty_bits(ctx.input_ty(insn, 0)) > 32 {
                     (ALUOp::SubS64, NarrowValueMode::ZeroExtend64)
                 } else {
@@ -1095,17 +1111,32 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
                     rm: zero_reg(),
                 });
                 Cond::Ne
-            } else {
-                let condcode = inst_condcode(ctx.data(insn)).unwrap();
-                let cond = lower_condcode(condcode);
-                let is_signed = condcode_is_signed(condcode);
-                // Verification ensures that the input is always a
-                // single-def ifcmp.
-                let ifcmp_insn = maybe_input_insn(ctx, inputs[0], Opcode::Ifcmp).unwrap();
-                lower_icmp_or_ifcmp_to_flags(ctx, ifcmp_insn, is_signed);
-                cond
             };
 
+            // csel.cond rd, rn, rm
+            let rd = get_output_reg(ctx, outputs[0]);
+            let rn = put_input_in_reg(ctx, inputs[1], NarrowValueMode::None);
+            let rm = put_input_in_reg(ctx, inputs[2], NarrowValueMode::None);
+            let ty = ctx.output_ty(insn, 0);
+            let bits = ty_bits(ty);
+            if ty_is_float(ty) && bits == 32 {
+                ctx.emit(Inst::FpuCSel32 { cond, rd, rn, rm });
+            } else if ty_is_float(ty) && bits == 64 {
+                ctx.emit(Inst::FpuCSel64 { cond, rd, rn, rm });
+            } else {
+                ctx.emit(Inst::CSel { cond, rd, rn, rm });
+            }
+        }
+
+        Opcode::Selectif | Opcode::SelectifSpectreGuard => {
+            let condcode = inst_condcode(ctx.data(insn)).unwrap();
+            let cond = lower_condcode(condcode);
+            let is_signed = condcode_is_signed(condcode);
+            // Verification ensures that the input is always a
+            // single-def ifcmp.
+            let ifcmp_insn = maybe_input_insn(ctx, inputs[0], Opcode::Ifcmp).unwrap();
+            lower_icmp_or_ifcmp_to_flags(ctx, ifcmp_insn, is_signed);
+
             // csel.COND rd, rn, rm
             let rd = get_output_reg(ctx, outputs[0]);
             let rn = put_input_in_reg(ctx, inputs[1], NarrowValueMode::None);

cranelift/filetests/filetests/vcode/aarch64/condops.clif

@@ -41,3 +41,14 @@ block0(v0: b1, v1: i8, v2: i8):
 
 ; check: subs wzr
 ; nextln: csel
+
+function %i(i32, i8, i8) -> i8 {
+block0(v0: i32, v1: i8, v2: i8):
+  v3 = iconst.i32 42
+  v4 = icmp.i32 eq v0, v3
+  v5 = select.i8 v4, v1, v2
+  return v5
+}
+
+; check: subs wzr, w0, #42
+; nextln: csel x0, x1, x2, eq